Sonnet 4.5 is the best coding model in the world

The model season is upon us. Nano this, mini that, 3.33 pro, banana, 6.1o ultra

can't wait for your video next week on gemini 3.0 (hopefully)

you should break out of the well-worn paths of your own benchmarks and standardized ways of testing coding agents and try A/B testing codex vs sonnet 4.5 in a fairly ambitious novel project and record your journey and lessons in prompting. you are a competent coder; there is a lack of good content with competent coders showing how to use agentic coders to mitigate their weaknesses and maximize their productivity while undertaking more ambitious projects (game design; complex webapps). something along those lines, at least.

Pretty sure that they just released Claude 4.5 because it was on there schedule and not because "they were scared of how good gpt5 codex is"

Still wondering when it gets cheaper

Instead of asking it to make a website from scratch to test the UI skills, ask it to implement a specific component or give it an already existing website and ask it to implement a new page following the same theme and general style of UI as the rest of the website.

Rumor has it that they written their articles with Claude.

This model is absolutely so much better than prior models from Anthropic. Been waiting for this

That part about your friend running a bunch of agents from his phone really got me hooked. Would love to see a video on that — maybe something like “Agent Farm” or “Agent Galley.”

love the long form content, keep up the great work Theo!